Skip to main content
AboutGuideContact
Sign in / Register

Image AI and Data

The Custom AI Model workspace teaches a detector from your own examples, then finds and counts that object across any image. Training runs in the cloud as a compute job, and detections render right on the image with counts you can review.

Custom AI Model

When auto-detection can't pick out the object you care about, a particular cell state, a morphological feature, or a marker arrangement, teach a model to find it. You draw a few examples, train, then run the model on new images. It all happens in one workspace, in three steps.

Placeholder: Screenshot Coming Soon

Custom AI Model workspace with example boxes drawn on an image and the label list in the sidebar

Draw examples

  1. Choose the images you'll teach from
  2. Drag a box around each example of your object, right on the image. Every box is a positive example of the same thing, so there's no per-box category to choose
  3. The sidebar lists what you've drawn and keeps a running count. Remove a stray box, or undo your last one
  4. Draw across as many images as you like. More representative examples help the model generalize

Tip: If your object is clearest in one fluorescence channel, switch the displayed channel (or step through Z-slices) while you draw. You're boxing what you see, so work in the channel where the object stands out.

Train

  1. Click Train Custom AI Model. Training is a paid feature, so on the free plan you'll see an upgrade prompt instead
  2. Training runs on the server and progress shows in real time
  3. When it finishes, your model is saved as a Model Card in the project's Models library

Training runs in the cloud and uses compute. See Image analysis compute for details.

Run

  1. Run your trained model on the image you're viewing, or across every image in the analysis at once
  2. Detections and a per-image count appear in the results gallery
  3. The model is reusable, so run it on new images anytime

Start from an Object Count

You don't have to label from scratch. In an Object Count analysis, after auto-detection, choose Train a Custom AI Model on these. Conspecta opens a new Custom AI Model analysis with those detections already drawn as editable boxes. Correct the ones that look wrong, add any it missed, and train. Your original Object Count analysis stays as it was.

The Model Library

Open the Models tab on the image-analysis landing page to see every custom model your project has trained. Each model is shown as a Model Card with:

  • A thumbnail mosaic of the training images
  • The model's name and training metadata
  • Quick actions to run inference, retrain, or rename

Model Cards are per-project, not shared across teams or other projects. If you want a model in another project, retrain it there.

Result Overlays

After a workflow or model run, results render directly on the image:

  • Center dots mark detected objects in the Object Count workflow
  • Density heatmap visualizes object density across the image
  • Threshold preview shows a binary mask of what passes the current threshold during workflow configuration

Available overlays depend on the workflow you ran.

The Data Tab

Switch from the image view to the Data tab to see every detected object in a sortable, filterable table.

Columns

  • Object ID within the image
  • Type indicating whether the row is a detection or a manual selection
  • Bounds (x, y, width, height) and center (x, y)
  • Area, perimeter, equivalent diameter (the diameter of a circle with the same area)
  • Feret diameter (minimum and maximum), the calipers of the object
  • Eccentricity, circularity (4π · area / perimeter²), aspect ratio
  • Average intensity across all channels, plus per-channel intensities for multi-channel images
  • Confidence when the row came from an AI model
  • Workflow-specific columns like marker classification for the Marker Classification workflow

Table Controls

  • Select individual rows or all rows, then hide, delete, or export
  • Open the Columns panel to toggle column visibility or reset to defaults
  • Sort by any column; filter by any value or range
  • Switch units between pixels and micrometers (calibrated from the image's scale metadata)

Summary Tab

The Summary tab shows aggregate statistics for the analysis (total counts, means, and workflow-specific metrics) in a one-glance view.

Export

Export results as Excel (.xlsx) from the Data tab. You can export all rows or just the selected ones. The export includes every column with values converted to your current measurement unit.

Training is server-side: Earlier versions of Conspecta trained models in the browser. Training now runs entirely on the server and uses compute, so your local machine no longer affects training speed.

Next Steps

Image viewingAnalyze flow cytometry